More Than Hallucinations: Five Developmental Processes at Stake in the Age of AI
By Jean Rhodes
Much of the public debate about artificial intelligence in higher education has focused on accuracy. Chatbots hallucinate. They generate plausible-sounding but factually incorrect answers about financial aid deadlines, course requirements, and mental health resources. They encode biases drawn from the data on which they were trained and can produce inconsistent responses to identical prompts. These are real and serious concerns, and institutions deserve the scrutiny they are receiving.
But there is a potentially more consequential concern that rarely makes headlines: the developmental costs to young people of replacing human relationships with machines. I recently published a paper in Applied Developmental Science on the developmental risks of AI in higher education, and what follows is the core of that argument. As I argue, the college years represent one of the most consequential periods in the human lifespan. Jeffrey Arnett described this stage as “emerging adulthood,” a time when young people gradually shift from dependence on family and institutions toward self-directed lives. This transition unfolds, crucially, through relationships with faculty, advisors, mentors, and other caring adults who know a student’s name, notice when something is wrong, and invest their scarce time and attention in ways that signal genuine regard. When colleges replace these relationships with chatbots, they are not merely cutting costs. They are intervening in a developmental process they do not fully understand and cannot fully control.
Identity. The college years are a critical period during which young people make commitments, explore those commitments in depth, and periodically revise them in response to feedback, friction, and disconfirming information from others. This requires encountering perspectives that challenge one’s own assumptions, hearing from a mentor that a career plan has blind spots, and sitting with uncertainty long enough to arrive at convictions that genuinely belong to oneself. Chatbots are largely designed to do the opposite. Optimized to maintain engagement and conversational flow, they tend to affirm users’ existing beliefs. The American Psychological Association recently warned that AI systems can function as digital echo chambers that reinforce users’ views even when those views are inaccurate. Experimental evidence suggests that even a single interaction with a sycophantic AI model can leave users more convinced they were right during interpersonal conflicts and less willing to apologize, repair relationships, or reconsider their own behavior. For students navigating a critical period of identity development, this is far from a neutral outcome.
Integrity. Chickering and Reisser identified integrity as a central developmental task of adulthood: the movement from externally imposed rules toward personally held values and a commitment to act consistently with them. When students routinely submit AI-generated work as their own, the self-narrative they construct increasingly rests on performances of competence rather than genuine accomplishment. Over time, the gap between apparent and actual achievement can erode confidence and authenticity. One recent study found that AI dependence predicted impostor syndrome symptoms among college students, with fear of negative evaluation serving as a key mechanism. Students may graduate with credentials that signal capabilities they do not genuinely feel they possess. The result is not simply weaker learning but a diminished sense of ownership over their own development.
Purpose. Purpose, the sense of direction, meaning, and long-term commitment that gives coherence to a life, develops through reflection and relationships. Research on narrative identity suggests that the stories people construct about who they are becoming constitute a primary pathway through which purpose emerges. This process depends on meaningful experiences and sustained relationships with adults who recognize students’ strengths and believe in their potential. When students rely on AI to interpret their experiences, draft personal statements, or work through difficult decisions, some of this interpretive work is outsourced. Sherry Turkle described the result as the “tethered self,” in which dependence on technology displaces inner dialogue and relational depth. A chatbot can simulate the appearance of helping a student think. It cannot participate in the developmental process through which purpose is actually formed.
Psychosocial development. Approximately half of all lifetime mental health disorders begin by age 18 and nearly two-thirds by age 25. The college years are therefore a period of genuine psychological vulnerability. Students with anxiety or obsessive-compulsive tendencies may find that chatbot use intensifies cycles of worry and reassurance-seeking. Those experiencing loneliness may come to rely on personalized, always-available AI instead of seeking human support. Recent research has revealed a troubling feedback loop in which depression predicts greater use of conversational AI for companionship, a relationship mediated by loneliness, such that students who are already isolated turn to chatbots for connection only to become more isolated over time. The developmental tasks of emerging adulthood, learning to tolerate uncertainty, regulate strong emotions, and seek support from others, are practiced through human relationships, not around them.
Social capital. Social capital refers to the resources, connections, information, and opportunities embedded in relationships. College is, among other things, a social networking institution. The introductions, referrals, letters of recommendation, and informal conversations that emerge from genuine mentoring relationships are not peripheral to the college experience. They are close to its core. The Gallup-Purdue Index found that graduates who strongly agreed that a professor cared about them as a person were nearly twice as likely to be engaged at work and satisfied with life after graduation. Since roughly half of jobs are secured through social networks, the replacement of mentoring relationships with chatbots carries serious long-term consequences for social mobility. These consequences are unlikely to be distributed equally. Economically advantaged students will continue to benefit from well-connected family members, friends, and mentors. Students from less privileged backgrounds, who may already hesitate to reach out for fear of confirming negative stereotypes, may increasingly be routed toward chatbots that deliver information without building relationships. The democratizing rhetoric that often surrounds AI in higher education deserves careful scrutiny.
Suggestions
None of this means AI has no place in student support. Colleges are genuinely stretched. Academic advising caseloads average roughly 300 students per advisor nationally and are often considerably higher. More than 60 percent of college students now meet criteria for at least one mental health problem. Campus counseling centers cannot keep pace. The question is not whether to use AI, but where to put it. In the ADS article, I argue for a human-at-the-helm model. In this approach, AI handles structured administrative work behind the scenes. It gathers notes, aggregates academic records, flags patterns worth discussing, and surfaces relevant resources so that advisors and mentors arrive at meetings more informed and less burdened. Studies of AI-assisted clinical documentation suggest that such systems can save practitioners roughly 20 percent of their time, hours that can be redirected toward the relational work that actually drives development. The human remains the point of contact, the interpreter, and the relationship. The AI stays in the background.
This model has implications that extend well beyond college advising offices. Formal mentoring programs serving youth and young adults face many of the same constraints: volunteer and staff capacity stretched thin, limited infrastructure for tracking mentee progress, and difficulty identifying at-risk youth before small problems become larger ones. A human-at-the-helm approach offers these programs a way to use AI to handle case documentation, surface conversation prompts grounded in evidence-based practice, and flag concerning patterns for program coordinators, all while keeping the mentor-mentee relationship firmly at the center. The goal in both settings is the same: to use technology to scale the administrative burden so that human attention can go where it matters most.
Colleges and mentoring programs alike are being asked to do more with less. AI can genuinely help with that, if it is positioned as a tool for supporting human relationships rather than substituting for them. If chatbots become the default source of guidance where human connection once existed, they will gradually lower young people’s expectations for what mentoring can be, training them to treat transactional exchanges as adequate substitutes for genuine developmental relationships. Each time a young person turns to a machine for reassurance, career guidance, or emotional support, the technology’s convincing simulation is reinforced while the human relationships essential to a flourishing life are quietly eroded. The hallucinations are the easy problem to fix. The developmental costs will take a generation to measure and much longer to repair.
